Is matrix multiplication multi-threaded for Double64?

I tested multiplication of two Double64 matrices and saw no effects on performance by using BLAS.set_num_threads(n).

Do you mean Float64? If so, yes. What size matrices?

No, I mean Double64 in DoubleFloats.jl.

It both isn’t threaded and it also isn’t using BLAS. BLAS only supports Float32 and Float64.

2 Likes

To be precise, it supports

julia> LinearAlgebra.BlasFloat
Union{Float32, Float64, ComplexF64, ComplexF32}
1 Like